Association of Medicare Mandatory Bundled Payment Reform With Joint Replacement Surgery Use for Beneficiaries With Alzheimer Disease and Related Dementias

Key Points

Question  Is the Comprehensive Care for Joint Replacement (CJR) model associated with changes in hip and knee replacement use for Medicare beneficiaries with Alzheimer disease and related dementias (ADRD)?

Findings  In this cohort study of 24 598 729 beneficiary-year observations among 9 624 461 unique beneficiaries, CJR was statistically significantly associated with a decrease in hip replacement use for beneficiaries with and without ADRD; however, the gap in use between these groups did not change with CJR implementation. The CJR model was not associated with changes in knee replacement use.

Meaning  This study found that the CJR model was not associated with a disproportionate reduction in joint replacement use for Medicare beneficiaries with ADRD.

Importance  Medicare beneficiaries with Alzheimer disease and related dementias (ADRD) are a particularly vulnerable group in whom arthritis is a frequently occurring comorbidity. Medicare’s mandatory bundled payment reform—the Comprehensive Care for Joint Replacement (CJR) model—was intended to improve quality and reduce spending in beneficiaries undergoing joint replacement surgical procedures for arthritis. In the absence of adjustment for clinical risk, hospitals may avoid performing elective joint replacements for beneficiaries with ADRD.

Objective  To evaluate the association of the CJR model with utilization of joint replacements for Medicare beneficiaries with ADRD.

Design, Setting, and Participants  This cohort study used national Medicare data from 2013 to 2017 and multivariable linear probability models and a triple differences estimation approach. Medicare beneficiaries with a diagnosis of arthritis were identified from 67 metropolitan statistical areas (MSAs) mandated to participate in CJR and 104 control MSAs. Data were analyzed from July 2020 to July 2021.

Exposures  Implementation of the CJR model in 2016.

Main Outcomes and Measures  Outcomes were separate binary indicators for whether or not a beneficiary underwent hip or knee replacement. Key independent variables were the MSA group, before-CJR and after-CJR phase, ADRD diagnosis, and their interactions. The linear probability models controlled for beneficiary characteristics, MSA fixed effects, and time trends.

Results  The study included 24 598 729 beneficiary-year observations for 9 624 461 unique beneficiaries, of which 250 168 beneficiaries underwent hip and 474 751 underwent knee replacements. The mean (SD) age of the 2013 cohort was 77.1 (7.9) years, 3 110 922 (66.4%) were women, 3 928 432 (83.8%) were non-Hispanic White, 792 707 (16.9%) were dually eligible for Medicaid, and 885 432 (18.9%) had an ADRD diagnosis. Before CJR implementation, joint replacement rates were lower among beneficiaries with ADRD (hip replacements: 0.38% vs 1.17% for beneficiaries with and without ADRD, respectively; P < .001; knee replacements: 0.70% vs 2.25%; P < .001). After controlling for relevant covariates, CJR was associated with a 0.07-percentage-point decline in hip replacements for beneficiaries with ADRD (95% CI, −0.13 to −0.001; P = .046) and a 0.07-percentage-point decline for beneficiaries without ADRD (95% CI, −0.12 to −0.02; P = .01) residing in CJR MSAs compared with beneficiaries in control MSAs. However, this change in hip replacement rates for beneficiaries with ADRD was not statistically significantly different from the change for beneficiaries without ADRD (percentage point difference: 0.01; 95% CI, −0.08 to 0.09; P = .88). No statistically significant changes in knee replacement rates were noted for beneficiaries with ADRD compared with those without ADRD with CJR implementation (percentage point difference: −0.03, 95% CI, −0.09 to 0.02; P = .27).

Conclusions and Relevance  In this cohort study of Medicare beneficiaries with arthritis, the CJR model was not associated with a decline in joint replacement utilization among beneficiaries with ADRD compared with beneficiaries without ADRD in the first 2 years of the program, thereby alleviating patient selection concerns.

Alzheimer disease and related dementias (ADRD) are among the leading causes of morbidity and mortality among older adults,¹ making the 6.2 million older individuals in the US with ADRD a particularly vulnerable and high-priority population.² The physical and cognitive impairments due to ADRD progressively increase and severely limit a patient’s ability to carry out routine activities, placing them at an increased risk of health care use and death and imposing an unprecedented demand on their families and caregivers. In 2021, Medicare alone financed more than 50% of the $355 billion that was spent on ADRD care, with hospitalizations, emergency department use, and postacute care being the main drivers of this spending.¹

Because increasing age is an important risk factor for ADRD³ and arthritis is a frequently occurring comorbidity,⁴ patients with ADRD are likely to need arthritis treatments, including the use of elective joint replacements (total hip [THR] and total knee replacements [TKR]). These surgeries are highly effective in alleviating pain, improving physical function, and enhancing health-related quality of life.⁵ Because joint replacements are among the most frequently performed inpatient surgical procedures for older Medicare beneficiaries,⁶ and there is considerable variability in the outcomes and spending for these procedures, joint replacements are included in several payment reforms launched by the Centers for Medicare & Medicaid Services (CMS). One such reform is the Comprehensive Care for Joint Replacement (CJR) model, which was mandated for most hospitals located in 67 randomly selected metropolitan statistical areas (MSAs) in 2016.⁷ The CJR bundles spending for the joint replacement episode, which includes the inpatient stay and 90-day postacute care period. Hospitals are accountable for maintaining the episode spending below the quality-adjusted spending benchmarks. Depending on their performance, hospitals can earn reconciliation payments or are required to repay CMS. Although the CJR model was found to modestly reduce joint replacement spending without compromising quality,⁸ it was also found to widen the gap in TKR use between White and Black Medicare beneficiaries.^9,10

The CJR model’s potential to limit access to joint replacements for vulnerable patients may be particularly relevant for beneficiaries with ADRD. Although the CJR model adjusts its spending benchmarks using the Medicare Severity Diagnosis Related Groups and diagnosis of fractures, it does not explicitly account for the clinical and social risk of patients. Hence, to achieve higher performance on CJR metrics, hospitals may selectively avoid elective surgery for beneficiaries with ADRD who have disproportionately higher clinical needs, a higher risk of adverse events, and increased spending. Thus, Medicare beneficiaries with ADRD, especially those with arthritis, are vulnerable and at an increased risk of being discriminated against or encountering barriers to the use of joint replacements compared with beneficiaries without ADRD.

The objective of our study was to examine the association of the CJR model with the utilization of elective THRs and TKRs for Medicare beneficiaries with and without ADRD. We hypothesize that a substantial gap exists in joint replacement use between beneficiaries with and without ADRD and that the CJR was associated with a greater decline in the use of these procedures for beneficiaries with ADRD, thereby widening the gap in joint replacement use between the 2 groups. Our study is the first to our knowledge to investigate whether mandatory bundled payment reforms such as the CJR model are likely to be associated with a decline in joint replacement use for Medicare beneficiaries with ADRD, thereby providing vital evidence regarding the role of such reforms in creating barriers to health care use and access for beneficiaries with ADRD.

We obtained demographic and enrollment data of all Medicare beneficiaries from the 2013 to 2017 Medicare Master Beneficiary Summary Files (MBSF).¹¹ We used these files to identify Medicare beneficiaries who met our inclusion criteria and lived in the 67 MSAs mandated to participate in the CJR program or the 104 control MSAs (eAppendix 1 in the Supplement). The MSAs were randomized to treatment and control groups by Medicare.⁷ Although 75 MSAs were originally assigned to the treatment group, 8 MSAs were excluded from participation at the start of the CJR implementation mostly owing to changes in participation in other payment reforms.¹² We used the MBSF-Chronic Conditions (MBSF-CC) segment to identify chronic conditions among Medicare beneficiaries. These chronic conditions are determined by Medicare using validated algorithms and claims data.¹³ We limited the cohort to beneficiaries who met the claims criteria for rheumatoid arthritis or osteoarthritis in a given data year because these beneficiaries are most likely to be at risk/eligible for THRs/TKRs.

The Medicare Provider Analysis and Review files contain details about inpatient and skilled nursing facility stays covered by Medicare.¹⁴ We used CJR-specific and other inclusion criteria to identify stays for elective THRs and TKRs (eAppendix 1 in the Supplement). We excluded hip fracture stays from the cohort because hospitals are unlikely to discriminate between patients in nonelective situations. We excluded stays in hospitals that were not reimbursed by the Inpatient Prospective Payment System and hospitals that participated in the Bundled Payments for Care Improvement program. The main analytic cohort consisted of 24 598 729 beneficiary-year observations from 2013 to 2017 for 9 624 461 unique beneficiaries with arthritis, of which 250 168 underwent THRs and 474 751 underwent TKRs. The study was approved and granted a waiver of participant informed consent by the University of Rochester Research Subject Review Board.

The outcomes were separate binary indicators for whether a beneficiary underwent a THR or TKR during a given year. The study had 3 key independent variables. First was a binary indicator of whether a beneficiary resided in an MSA that was mandated to participate in the CJR or in a control MSA. The second variable was the before-CJR or after-CJR phase with the before-CJR phase extending from 2013 to 2015 and the after-CJR phase being 2017. We excluded observations from 2016 in the main multivariable analysis because CJR’s implementation in April 2016 does not permit the classification of beneficiaries in 2016 into a before-CJR or after-CJR cohort. Third was a binary indicator of whether or not a beneficiary was ever diagnosed with ADRD.^15-17 This indicator was determined by Medicare using ADRD diagnosis codes in multiple claims data sources. In statistical analysis, we included the main effects of these 3 variables along with their 2-way and 3-way interactions.

The multivariable models controlled for continuous specifications of age and calendar year and categorical specifications of sex, race and ethnicity (Asian, Hispanic, non-Hispanic Black, non-Hispanic White, North American Native, unknown, other), dual eligibility, and 23 chronic conditions. Race and ethnicity data were obtained from the MBSF, which is populated from Social Security Administration records.¹⁸

We report means (and SDs) for continuous variables and numbers (and percentages) for categorical variables. We used Kruskal-Wallis and χ² tests to test differences in the distribution of variables across treatment and control MSAs. We estimated multivariable linear probability models with MSA-level fixed effects and Huber-White robust sandwich estimators of variance. These models tested whether CJR was associated with higher or lower THR/TKR use for beneficiaries with ADRD compared with those without ADRD. We used a triple differences estimation approach to determine this association.^9,19 Before estimating these models, we tested for the parallel trends assumption, which examines trends in surgical use for beneficiaries with and without ADRD before CJR implementation. In case of violation of the assumption, we included interactions of the year with CJR treatment/control MSA and ADRD indicators. We used Lewin Group’s methods to determine MSA-level weights, which accounted for selection probabilities of the treatment and control MSAs.

All analyses were performed using Stata/MP version 16.1 (Unix) (StataCorp LLC). A 2-tailed P < .05 was considered statistically significant. Details of the analytic and weighting approaches and the null and alternate hypotheses are provided in eAppendixes 2 and 3 in the Supplement.

First, we included all eligible beneficiaries and did not limit the cohort to beneficiaries with arthritis. Second, we reestimated the models using observations from the 75 treatment MSAs originally mandated to participate in the CJR program (intention-to-treat analysis) and the corresponding 121 control MSAs. Third, we defined elective surgical procedures using Medicare’s criteria for calculating risk-standardized complication and readmission rates for joint replacements.²⁰ Fourth, we limited the cohort to beneficiaries diagnosed with ADRD within the past year because these beneficiaries are likely to derive greater value from joint replacements than those with long-standing ADRD. Fifth, we reestimated the models among those with Alzheimer disease only instead of those with ADRD. Sixth, we estimated logistic regression models instead of linear probability models to determine the sensitivity of the main findings to model specifications. Finally, we included observations from 2016 into the after-CJR phase because hospitals may have introduced changes before the start of the CJR in April 2016.

We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline to report this study.²¹ Data were analyzed from July 2020 to July 2021.

The cohort in 2013 included 4 688 663 Medicare beneficiaries from 67 CJR MSAs and 104 control MSAs (Table 1). Of these beneficiaries, 885 432 (18.9%) had a diagnosis of ADRD, with ADRD prevalence being 20.1% in CJR MSAs and 17.9% in control MSAs (P < .001). Overall, the mean (SD) age was 77.1 (7.9) years, 3 110 922 (66.4%) were women, 3 928 432 (83.8%) were non-Hispanic White, and 792 707 (16.9%) were dually eligible for Medicaid. The mean (SD) number of comorbidities (excluding ADRD) was 4.4 (2.5).

In 2013, the overall THR rate was 1.0%, with the rate being lower in CJR MSAs compared with control MSAs (1.0% vs 1.0%; P < .001) (Table 1). The overall TKR rate was 2.1%, with the rate being lower in CJR MSAs compared with control MSAs (2.0% vs 2.2%; P < .001). Additional descriptive statistics comparing cohorts in CJR and non-CJR MSAs are presented in eTables 1-3 in the Supplement, and eTable 4 in the Supplement compares beneficiaries with and without ADRD.

Before CJR implementation, THR rates for beneficiaries with and without ADRD were 0.38% (95% CI, 0.36% to 0.39%) and 1.17% (95% CI, 1.16% to 1.18%), respectively, in CJR MSAs (Figure 1 and Table 2, column A). This THR rate for beneficiaries with ADRD was 0.79 percentage points lower (95% CI, −0.81 to −0.77; P < .001) compared with the rate for beneficiaries without ADRD. Similarly, TKR rates for beneficiaries with and without ADRD were 0.70% (95% CI, 0.67% to 0.72%) and 2.25% (95% CI, 2.23% to 2.26%), respectively, in CJR MSAs. This TKR rate for beneficiaries with ADRD was 1.55 percentage points lower (95% CI, −1.58 to −1.52; P < .001) than beneficiaries without ADRD. Similar gaps in surgery use were noted in the period after CJR implementation in CJR MSAs, and among control MSAs.

The parallel trends assumption was violated only for the THR cohort (eTable 5 in the Supplement). After controlling for relevant covariates, the CJR model was associated with a 0.07-percentage-point decline in THR use for beneficiaries with ADRD (95% CI, −0.13 to −0.001; P = .046) and a 0.07-percentage-point decline for beneficiaries without ADRD (95% CI, −0.12 to −0.02; P = .01) residing in CJR MSAs compared with beneficiaries residing in control MSAs (Table 2, column C). However, the decline for beneficiaries with ADRD was not statistically significantly different from the decline for beneficiaries without ADRD (percentage point difference, 0.01; 95% CI, −0.08 to 0.09; P = .88) (Table 2, column D), indicating that the gap in THR use between these 2 groups did not change with CJR implementation. Despite this statistically nonsignificant finding, the change in THR use was consistent with effect sizes ranging from a −0.08-percentage-point difference (decline in THR use) to a 0.09-percentage-point difference (increase in THR use) for Medicare beneficiaries with ADRD compared with those without ADRD.

The CJR model was not associated with statistically significant changes in TKR use for beneficiaries with (percentage point change, −0.01; 95% CI, −0.05 to 0.04; P = .78) or without ADRD (percentage point change, 0.02; 95% CI, −0.01 to 0.06; P = .15). The change for beneficiaries with ADRD did not statistically significantly differ from the change for beneficiaries without ADRD (percentage point difference, −0.03; 95% CI, −0.09 to 0.02; P = .27). Despite this statistically nonsignificant finding, the change in TKR use was consistent with effect sizes ranging from a −0.09-percentage-point difference (decline in TKR use) to a 0.02-percentage-point difference (increase in TKR use) for Medicare beneficiaries with ADRD compared with those without ADRD. The adjusted changes in rates are graphically presented in Figure 2, and the regression estimates are presented in eTable 6 in the Supplement.

The inferences from the sensitivity analysis were consistent with the main analysis (eTable 7 in the Supplement). An exception was that in the analysis limiting the ADRD cohort to beneficiaries diagnosed with ADRD in the past year, CJR was found to worsen the gap in TKR use between beneficiary groups (percentage point difference, −0.09; 95% CI, −0.18 to 0.00; P = .05).

In this national analysis of Medicare beneficiaries from 2013 to 2017, we found that Medicare’s mandatory bundled payment reform, the CJR model, was associated with an overall decline in THR use for both beneficiaries with and without ADRD. However, the decline for beneficiaries with ADRD was not statistically significantly different from that for beneficiaries without ADRD, indicating that the gap in THR use between the 2 groups remained unchanged 2 years after CJR implementation. We did not find statistically significant changes in TKR use with CJR implementation, nor did the CJR widen the preexisting gap in TKR use between beneficiaries with and without ADRD. In a sensitivity analysis limited to beneficiaries diagnosed with ADRD in the past year, the CJR model was associated with a decline in TKR use for these beneficiaries. However, these findings need to be interpreted with caution and should be viewed as hypothesis generating.

Recent studies have shown that CJR may incentivize hospitals to selectively avoid higher-risk beneficiaries because of the lack of clinical and social risk adjustment.^9,10 This selection process may result in vulnerable beneficiaries such as racial and ethnic minority groups and low-income beneficiaries facing increasing barriers to undergo these much-needed surgeries. In theory, similar barriers may arise for beneficiaries with ADRD. First, because beneficiaries with ADRD have more significant physical and cognitive deficits, these patients are more likely to require costly institutional postacute care, such as skilled nursing facilities,^22,23 further increasing spending. Because hospitals in the CJR model are known to control spending primarily by discharging patients to home instead of skilled nursing facilities,^8,24 this strategy in particular may place patients with ADRD at an increased risk of complications and suboptimal outcomes, further discouraging hospitals and surgeons from operating on patients with ADRD. Second, patients with ADRD undergoing joint replacements have higher rates of mortality^25,26 and complications such as recurrent dislocation^27,28 or delirium^29,30 compared with patients without ADRD. These complications and associated readmissions are likely to be adversely associated with hospital quality and spending performance in the CJR model. Given these mechanisms, our finding that the CJR model did not disproportionately worsen joint replacement use for beneficiaries with ADRD is reassuring.

The present study found that joint replacement use was statistically significantly lower for beneficiaries with ADRD than for those without ADRD. This difference may result from the weakened physical and cognitive status of beneficiaries with ADRD, which places them at an increased risk of adverse events, mortality, and long-term care needs.^25,26 This increased risk is likely to contribute to decisions by patients, family members, and physicians to avoid surgery. The finding of declines in joint replacement use for both groups of beneficiaries following CJR implementation may result from a careful selection of patients such that quality and spending scores for hospitals are optimized. This finding differs from previous research that examined joint replacement volume as a secondary outcome and did not find statistically significant changes with CJR implementation,^8,24 and from previous conceptual frameworks that bundled payments may increase hospital joint replacement volume because of increased efficiency resulting from CJR-motivated quality improvement initiatives.^31,32 The difference in findings is likely explained by the cohort definition, our study design focusing on beneficiary-level instead of aggregate MSA-level joint replacement use, and our granular adjustment for comorbidities. Our finding of a decline in TKR use for beneficiaries recently diagnosed with ADRD is concerning and needs to be further investigated because these surgical procedures may have the greatest value for individuals who are in the early stages of the ADRD disease process.

The study findings have important policy, practice, and research implications. First, in recent years there has been a growing call for CMS to consider more comprehensive risk adjustment of reforms to account for the increased clinical and social risk of vulnerable beneficiaries, such as those with ADRD.³³ A recent update to the CJR model will now result in adjusting the benchmarks for age, a Hierarchical Condition Category measure of clinical risk, and dual eligibility for Medicaid.⁷ This may alleviate the overall decline in THR use and the decline in TKR use for beneficiaries newly diagnosed with ADRD that we note. Second, although the CJR model and other payment reforms are being evaluated for their potential unintended effects on surgery use among racial and ethnic minority groups and low-income beneficiaries,^9,10 little attention has been paid to other vulnerable beneficiaries, such as those with ADRD. A greater emphasis on this patient population will ensure that reforms do not create barriers to health care use for those in need of care. Third, the study’s population-level findings are driven by granular decisions made by surgeons and other clinicians. A continual examination of the clinical decision-making process, including goals of care and patient preferences, will be needed to determine the appropriateness of surgery under different scenarios. Finally, further work is needed to monitor and correct the unintended yet likely consequences of not only the CJR but also of other payment reforms.

The present study has several limitations. First, the analysis does not account for unmeasured confounders that may influence the decision to undergo surgery, such as patient preference for surgery or availability of a caregiver, because this information is not available in national Medicare data. Second, we identified the presence of ADRD and arthritis using data from the MBSF-Chronic Conditions segment. Medicare uses rigorously validated algorithms to identify these conditions. However, because of the nature of the data, we were unable to identify the severity or progression of ADRD, and the severity, type, or location of arthritis. These concerns are partly mitigated by reestimating our models among beneficiaries who were newly diagnosed with ADRD and by using data from all Medicare beneficiaries in the sensitivity analysis. Moreover, the prevalence of these conditions is unlikely to change differentially over the study period. Finally, in 2018, the design of the CJR was changed from a fully mandated program to a partially mandated program. Because this program update changes the randomized selection of the MSAs, we limited the analysis to 2017, and we are unable to detect changes that may have occurred more than 2 years after implementation.

In this cohort study of Medicare beneficiaries with arthritis, we found that the CJR was not associated with a worsening of the gap in joint replacement use between Medicare beneficiaries with and without ADRD. With the onset of the value-based payment approach to health care, CMS has increasingly relied on reforms such as the CJR to improve the quality of care while controlling spending. Despite being well intended, these reforms may unintentionally limit care for vulnerable patient populations. Although the study did not find these unintended effects among beneficiaries with ADRD in the first 2 years of the CJR, continual monitoring of these reforms will ensure that beneficiaries with ADRD and other vulnerable patients receive equitable and effective care.

Accepted for Publication: December 15, 2021.

Published: February 11, 2022. doi:10.1001/jamahealthforum.2021.5111

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2022 Thirukumaran CP et al. JAMA Health Forum.

Corresponding Author: Caroline P. Thirukumaran, MBBS, MHA, PhD, Department of Orthopaedics, University of Rochester, 601 Elmwood Ave, PO Box 665, Rochester, NY 14642 (caroline_thirukumaran@urmc.rochester.edu).

Author Contributions: Dr Thirukumaran had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Thirukumaran, Li, Glance.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Thirukumaran.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Thirukumaran, Cai, Glance.

Obtained funding: Thirukumaran, Li.

Administrative, technical, or material support: Thirukumaran, Li.

Supervision: Ricciardi, Li.

Conflict of Interest Disclosures: Dr Thirukumaran reported receiving grants from the National Institute on Aging (administrative supplement) and the National Institute on Minority Health and Health Disparities during the conduct of the study. Dr Ricciardi reported receiving personal fees from DePuy Synthes and grants from Johnson & Johnson outside the submitted work. Drs Cai and Glance reported receiving grants from the National Institutes of Health during the conduct of the study. No other disclosures were reported.

Funding/Support: The authors gratefully acknowledge funding support from the National Institute on Aging under Award No. R01MD012422-S1, and the National Institute on Minority Health and Health Disparities of the National Institutes of Health under Award No. R01MD012422.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.